DexForce unveils UNITAS, the first 3D-native world action model, hitting 99.8% on LIBERO with just 1.7B params
jiqizhixin · x · 2026-10-19
DexForce and collaborators (CUHK-Shenzhen, Sun Yat-sen, etc.) released UNITAS, billed as the first 3D-native world action model for embodied manipulation (arXiv 2610.12099).
- Core argument: existing world action models translate between image, action, and world-change feature spaces, wasting capacity on "translation" rather than understanding — the same pixel shift doesn't mean the same physical distance across camera views.
- Previous methods are "3D-aware" (3D as auxiliary); UNITAS uses 3D physical space as the base representation for perception, action, and world change, removing cross-space translation overhead.
- Results: with only 1.7B parameters, UNITAS achieves 99.8% average success on LIBERO, 89.1% on LIBERO-Plus, and strong results on RoboTwin 2.0 dual-arm manipulation.
- Code and checkpoints are open-sourced; dataset coming soon.
More from Embodied
- Robots ace blind precision assembly using only joint encoder signals, no cameras — kevin_zakka · 2026-10-19
- If AI records every conversation by default, society changes dramatically — generativist · 2026-10-19
- Tesla shareholder letter teases a mysterious Optimus v3 silhouette — CyberRobooo · 2026-10-19
- Robotics is having its GPT-2 moment, says State of AI's Nathan Benaich — nathanbenaich · 2026-10-19
- Robotics founder's China day 6: not the top AI innovator, but adoption will lead by a massive gap — DominiqueCAPaul · 2026-10-19
- Ropedia Kicks Off Super Week With W0, a World Model Teaching Robots to Predict Physics Before Acting — liuziwei7 · 2026-10-19